ENH: add quantile method to resample

[jorisvandenbossche]

Problem description

The quantile method is currently not supported on resample. While resample(...).agg(lambda x: x.quantile(..)) works fine, it would be nice to add the resample(..).quantile(..) shortcut.

Code Sample

In [19]: ts = pd.Series(range(20), index=pd.date_range('2016-01-01', periods=20))

In [21]: ts.resample('W').agg(lambda x: x.quantile(0.75))
Out[21]: 
2016-01-03     1.5
2016-01-10     7.5
2016-01-17    14.5
2016-01-24    18.5
Freq: W-SUN, dtype: float64

So it would be nice the following would give the same result:

In [22]: ts.resample('W').quantile(0.75)
/home/joris/miniconda3/envs/dev/bin/ipython:1: FutureWarning: 
.resample() is now a deferred operation
You called quantile(...) on this deferred object which materialized it into a series
by implicitly taking the mean.  Use .resample(...).mean() instead
  #!/home/joris/miniconda3/envs/dev/bin/python
Out[22]: 14.25

The fact that this currently implicitly takes the mean before calculating the quantile (ts.resample('W').mean().quantile(0.75)) would make this change slightly API breaking.

Using pandas master, 0.19.0+289.g1bf94c8

[jorisvandenbossche]

BTW, using grouby this also works (and on rolling works as well):

In [23]: ts.groupby(pd.Grouper(freq='W')).quantile(0.75)
Out[23]: 
2016-01-03     1.5
2016-01-10     7.5
2016-01-17    14.5
2016-01-24    18.5
Freq: W-SUN, dtype: float64

In general should maybe check whether there are also other methods not working on resample.

[jreback]

yeah, we have specificially defined Resampler methods (as we do for groupby).

[srinivasreddy]

I am taking it

[jorisvandenbossche]

@jreback can you remember if there was a specific reason that quantile was left out? or just an oversight?

[jreback]

IIRC, some methods (e.g. cython defined ones, and some others), are specifically defined in .groupby, so these are 'named' methods and were transfered to .resample. However other methods are dispatched (IOW, on the whitelist but not defined as above). I don't think I audited all of these for suitability.

So in theory, we should evaluate these and define if appropriate (or possibly make a general method to make these definitions).